Why the default AI output fights your separation
Ask Midjourney, ChatGPT or Stable Diffusion for "a wolf illustration" with no other guidance, and you get a photographic-looking render: soft lighting, smooth gradients, hundreds of subtly different shades blending into one another. That is exactly the wrong shape of file for spot colour separation.
The separation tool sorts pixels into ink groups by colour family — hue and saturation, not just brightness — so it can tell a mid-tone of blue from a mid-tone of grey even though a brightness-only tool would confuse them. What it cannot do is guess how many inks you meant to use. Feed it a smooth gradient across fifty shades of blue and it will either flatten them into fewer inks than you wanted, or split them in places you did not expect. Ask the generator for flat colour instead, and the guesswork disappears before it starts.
Lino and block printing has the same problem for a different reason: there are no halftone dots in relief printing at all, so any shading in the source image has to become a flat shape at some cutoff point. An AI generator that hands you soft airbrushed shadow will not tell you where that cutoff should fall — you decide that when you prompt, not after you have already carved.
Prompting for halftone screening (matched to your mesh)
Halftone screening can absolutely hold soft shading — that is what it is for. The limit is not tone, it is the smallest dot your mesh can survive at wash-out. A coarse mesh needs bolder, simpler artwork; a fine mesh can hold real detail.
| Mesh | Smallest dot | Add to your prompt | Avoid |
|---|---|---|---|
| 110 | 0.46 mm (0.018″) | bold flat shapes, thick outlines, 2–3 colours, minimal detail | fine hairlines, small type, delicate texture |
| 156 | 0.28 mm (0.011″) | flat illustration, clean edges, 3–4 colours, moderate detail | photorealistic rendering, dense fine texture |
| 230 | 0.19 mm (0.0075″) | flat illustration with soft shading inside each shape, fine linework okay | smooth full-image gradients across many hues |
These are the same mesh presets the tool ships with — see the mesh count to LPI chart for where the numbers come from. A detail smaller than the smallest-dot figure for your mesh will not survive wash-out, no matter how sharp the AI render looked on screen.
Style words that consistently help: flat vector illustration, bold poster design, screen print style, limited colour palette, high contrast, solid shapes, thick outlines. Style words that consistently hurt: photorealistic, soft lighting, intricate detail, fine texture, smooth gradient.
Two example prompts that separate cleanly:
- "A howling wolf, flat 3-colour illustration, thick black outlines, bold poster style, high contrast, no gradients, plain background" — built for 110–156 mesh.
- "A botanical fern illustration, flat vector style, 4 spot colours, soft shading within each leaf, clean hard edges between colours, screen print poster art" — built for 230 mesh, where the finer dot can hold the shading.
Prompting for lino and block printing
Relief printing needs the opposite instruction. A roller lays down solid, opaque ink from a raised surface — there is no such thing as 40% ink from a lino block — so every tone in the source image has to resolve to a flat shape before you carve, not after.
Ask directly for a linocut style or woodblock print style: high-contrast black and white, bold negative space, thick clean lines, no gradients, no soft shadow. Most AI generators recognise "linocut" and "woodcut" as style terms and will bias toward flat, carved-looking shapes on their own, which is most of the work done for you.
| Add to your prompt | Avoid |
|---|---|
| linocut style, woodblock print, bold negative space | watercolour, soft shadow, airbrush |
| high contrast black and white, thick clean lines | photographic, gradient, soft focus |
| flat colour shapes, one tone per block, no shading | subtle tonal variation, fine crosshatching |
Planning a multi-block or reduction print? Ask for a specific number of tones — "3-tone illustration, light, mid, dark, each tone a separate flat shape" — and treat the result as a starting sketch, not a final file. No generator reliably delivers exactly three clean tones on the first try; expect to touch up boundaries in an image editor before you separate, the same way you would clean up a hand-drawn sketch.
One thing prompting cannot fix: isolated thin lines and floating islands. If your design uses hand-cut stencils or has fine bridges holding a shape together, check them visually after generating — an AI image has no idea a shape needs to stay physically connected to survive being cut. The lino and block printing guide covers island bridges, mirroring and registration once you have a clean flat-shape file.
After the image comes out
Three checks before you separate, in order:
- Export as PNG, not JPEG. JPEG compression blurs colour boundaries, and those blended edge pixels get read as a colour you never asked for during separation.
- Check the resolution. Aim for 300 PPI at your final print size — a 10-inch print wants roughly 3000 pixels on that side. Most generators produce far less natively; upscale before separating if needed, not after.
- Count the actual colours. Open the file and look for how many distinct colour regions it really has, not how many you asked for. AI generators drift: a "3-colour" prompt often ships with a fourth colour hiding in an edge or a shadow. Merge it in the tool's ink preview, or regenerate with a tighter prompt.
Got a flat-colour AI image ready?
Load it, set your ink count and mesh, and see the separation and the halftone preview instantly — before you commit a film or a block to it. Free while in beta.
A short checklist
Ask for
- Flat vector illustration or linocut/woodblock style
- A specific number of colours, matched to your ink count
- High contrast, bold outlines, solid shapes
- Soft shading only if your mesh is fine enough to hold it
Avoid asking for
- Photorealistic, smooth gradient, soft lighting
- Fine texture or intricate detail below your mesh's smallest dot
- Any shading at all, if the destination is a lino block
Common questions
Can I use AI-generated art for screen printing?
Yes, as long as it separates into a manageable number of flat, contrasting colours. Prompt for a flat illustration or poster style with a stated colour count rather than a photorealistic render, and check the actual colour count once it comes out before you separate it.
Which AI image generator works best for print-ready art?
The prompt wording matters more than the tool. Any generator that responds to style terms like "flat vector illustration", "screen print poster" or "linocut" will get you most of the way there — Midjourney, ChatGPT and Stable Diffusion all recognise those terms.
Why does my AI image separate into more colours than I asked for?
Antialiasing and soft gradient noise that the generator adds even to a "flat" image get read as extra colours during separation. Merge the extra ink in the tool's preview, or regenerate asking explicitly for no anti-aliasing and no gradients, and export as PNG rather than JPEG.
Does my mesh count change how I should prompt?
Yes. A coarse mesh like 110 needs bold, simple shapes with almost no fine detail; a fine mesh like 230 can hold soft shading and finer linework. See the mesh table above for the smallest dot each one can survive at wash-out.
Can I use the same AI image for lino as for screen printing?
Usually not without changing the prompt. Lino needs every tone reduced to a flat shape before you carve, since there is no halftone step in relief printing. Screen printing can hold soft shading within one ink. Prompt for the technique you are actually using.
What resolution should I generate or upscale to?
300 PPI at your final print size. A 10-inch print needs roughly 3000 pixels on that side. Most AI generators produce far less natively, so upscale before you separate, not after.
We are new, and reviews, testers and ideas make this better. Write a review · Become a tester · Send an idea · Show us your work